Hi,

We use Windows Magic Professional to monitor our data. Each time the data for a sample is collected, the software does the tests for special causes and plots the point on the X-bar and R control chart.

Say I have collected data for the 1st three samples and the 3rd sample is out of control. Since the control limits are calculated for every sample. Now, say I have collected data for samples 4, 5, 6, 7 and none of them were out of control but now my 3rd sample which was out of control shows it's in control. So I want to know whether I should monitor the data continuously for each sample or after every lot?

The drawback with this software is, it does the tests for special causes only on the last sample. Say you've collected 8 samples, it does all the tests only on the 8th sample and when the data for 9th sample is collected, it does the tests only for the 9th sample.

So can someone tell me what to do?

The issue is that you are allowing the software to recalculate the control limits with the addition of every subgroup. This is not how SPC is intended to work.

One should collect an appropriate number of subgroups (based on knowledge of how the process actually varies). The rule of thumb fo rthis is 20-25 subgroups but may be more or less depending on the process.

Calculate the control limits and adjust for any obvious assignable cause results. Apply these limits to all future subgroups.

Only recalculate the limts when an intentional permanent improvement is made.

So do I need to fix my control limits, in order to not recalculate it?

But I thought we don't set any control limits.

I use Minitab but since it isn't a real time software, I always interpreted the charts after all the samples were collected. But with the real time software I don't know whether I had to taken any action in between, if it is out-of-control. Or wait until the end of each lot. In most of our processes we collect around 17 samples with a sample size of 3.

We use Windows Magic Professional to monitor our data.
What software? I can't find any SPC software called Windows Magic Professional.

Unfortunately they don't even have a website. We bought it 2001.
I know the software has drawbacks. I got the name of the software wrong. This is the address

Magic Windows Professional
The Crosby Company
P.O Box 188
Wedron, IL 60557
www.qualitynews.com (http://www.qualitynews.com)

I agree with Bev D. The control limits should be calculated after the initial study and should not recalculated based on every subsequent sample unless something is corrected in the process.

In my opinion, it is also appropriate after an extensive production run, and after all data points with special/assignable causes are removed, to do a recalculation of the limits.

Just for the reasons you mentioned, constant recalculations will mask any trends or out of control issues. If you have a single sample that is out of control, and several following samples are then in control, I would reverify the measurement on the one piece (if possible) to ensure there was no measurement or recording error. Based on the fact the process stayed in control after that one instance, I would probably assume that the one sample was indeed a special cause.

One thing I really don't understand is how a software program can make the determination (based on the latest data point) that a piece had special cause variation without further analysis of the process?

I think the OP raises a big issue. How many organizations are applying SPC blindly without fully understand how to use the software or the concepts. In 2001 when the software was purchased, what was its intended purpose? Eight years later, control limits are still being dynamically calculated. Sounds like for 8 years, they misapplied SPC.

I think the OP raises a big issue. How many organizations are applying SPC blindly without fully understand how to use the software or the concepts. .... Sounds like for 8 years, they misapplied SPC.

:agree1: Yes, we know that blindly rubber stamping X-bar-R charts alone may have been misapplied SPC.

:agree1: Yes, we know that blindly rubber stamping X-bar-R charts alone may have been misapplied SPC.
Unfortunately there are lots more people using SPC without understanding even basic principles than there are people who do understand what they're doing. The community of happily ignorant users increased dramatically with the advent of Six Sigma, with many people being told that they don't need to understand what they're doing so long as they can plug the numbers in and see a result that matches what they're looking for. And please, no responses about how a true SS practitioner would never do or advocate such thing, which is a tautological concept.

Unfortunately there are lots more people using SPC without understanding even basic principles than there are people who do understand what they're doing.

It has been a double edge sword. In my mind, in many ways automotive is to blame by forcing SPC and capability requirements onto the entire industry without the proper depth of understanding in place beforehand - even for the OEMs! I will say that the AIAG books have been some of the better attempts to cover the topics...as a starting place, not as a final rule.

Yet, the other industries that have been slow to incorporate SPC and process capability are really spinning their wheels. I have been there, and it is so old school it makes the hair stand up on my neck.

On top of that, I think SPC is - and should be - an evolving developmental process. As we slowly pry ourselves away from plug-and-chug rubber stamping and start to think about each and every process in a more holistic approach, we will learn new and effective techniques for control. The job is far from done, and the harm of rubber stamping is far from being realized and reconciled.

But, we keep trying!

It has been a double edge sword. In my mind, in many ways automotive is to blame by forcing SPC and capability requirements onto the entire industry without the proper depth of understanding in place beforehand - even for the OEMs! I will say that the AIAG books have been some of the better attempts to cover the topics...as a starting place, not as a final rule.

Yet, the industries that have been slow to incorporate SPC and process capability are really spinning their wheels. I have been there, and it is so old school it makes the hair stand up on my neck.

On top of that, I think SPC is - and should be - an evolving developmental process. As we slowly pry ourselves away from plug-and-chug rubber stamping and start to think about each and every process in a more holistic approach, we will learn new and effective techniques for control. The job is far from done, and the harm of rubber stamping is far from being realized and reconciled.

But, we keep trying!

I've always thought that part of the reason for the lack of understanding of statistical principles lies in the idea that it's seen as a "control method" when it's really about confirmation of the efficacy of control methods. You start with a hypothesis that says, "If I control [process variables] x, y and z I should see continuous conforming output" and then test the hypothesis using the proper statistical tools. It's really just the scientific method, and it might be found that confirmation of the hypothesis means that ongoing SPC isn't even necessary.

I've always thought that part of the reason for the lack of understanding of statistical principles lies in the idea that it's seen as a "control method" when it's really about confirmation of the efficacy of control methods.

Some of the tools are confirmation, but others have value in their predictive capability - again, only when used correctly. When use incorrectly, the true benefit is masked.

SPC as "report card charting" is a good example of confirmation of the efficacy of control methods. It is a use, although one of the weaker ones.

But, using SPC as a tool to tell you when something is occurring, and that if you adjust the process in time you can prevent bad product from being made, then it is part of the control. Much like the lines on the road, they don't tell you that you have driven well, but if you cross them they provide you a signal that you should either readjust, or you will be "out of specification".


You start with a hypothesis that says, "If I control [process variables] x, y and z I should see continuous conforming output" and then test the hypothesis using the proper statistical tools.

Sure, it is one use of statistics - more along the lines of the result of design of experiments - another statistical tool that is not control.

It's really just the scientific method, and it might be found that confirmation of the hypothesis means that ongoing SPC isn't even necessary.

As long as your population studied is completely represented by your sampling, perhaps so - which would mean under all conditions it is impossible to make bad product. I'd buy that....if I ever saw it. Customers may be skeptical, though. That is how they are.

Some of the tools are confirmation, but others have value in their predictive capability - again, only when used correctly. When use incorrectly, the true benefit is masked.
If statistics can't be use for prediction, there's no point in using statistics. The idea of confirmation is a way of predicting the future performance of a process by controlling the appropriate variables. The null hypothesis is a a form of prediction that's confirmed or disconfirmed by the test(s).

SPC as "report card charting" is a good example of confirmation of the efficacy of control methods. It is a use, although one of the weaker ones. It's not "report card" anything, nor is it comparatively weak. Ongoing monitoring/measurement is almost always necessary, but that doesn't mean that ongoing charting is always necessary. There are times when, as Bobby Zimmerman said, you don't need a weatherman to know which way the wind blows.

But, using SPC as a tool to tell you when something is occurring, and that if you adjust the process in time you can prevent bad product from being made, then it is part of the control.
There are times (machining) when ongoing charting can be helpful in the way you describe, but even then it depends on the factors involved. Sometimes variables can't be controlled to the extent that would be necessary to abandon the control charts--intrinsic differences between lots of raw materials is one example.


As long as your population studied is completely represented by your sampling, perhaps so - which would mean under all conditions it is impossible to make bad product. I'd buy that....if I ever saw it. Customers may be skeptical, though. That is how they are.
Like I said, it depends on the extent of control over variables that can be invoked, as well as the interactions of variables. Sometimes there's so much s*** going on that charting is your only hope. In other cases, not so much, and ongoing charting is a waste of time.

Like I said, it depends on the extent of control over variables that can be invoked, as well as the interactions of variables. Sometimes there's so much s*** going on that charting is your only hope. In other cases, not so much, and ongoing charting is a waste of time.

Conversely, sometimes there's so much s*** going on that charting is nearly hopeless - like injection molding.

It's not "report card" anything, nor is it comparatively weak.

Well, leave it to customers to demand things that have little value, there is "report card" charting. And, it is weak. The process may be controlled (by SPC or whatever is appropriate...) by a process parameter - such as pressure or wattage, but the customer wants to see a chart of the resulting print characteristic, anyway. No predictive value - the deed was already done.

Statistics as a tool when applied correctly can be predictive. The problem is most SS practioners and rogue statisticians apply a "formula" to get a result, then try to predict future results without any understanding of the variability of the result.

It is like saying because you won the first game of the baseball season you will win 162 games. As more data is collected, the variability stabilizes enough to be predictive.

If there is too much going on in the process, DOE should be applied to separate out signal from noise, but of course that does not get applied correctly either.